1628
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 6, NOVEMBER/DECEMBER 2000
A Neural-Network-Based Space-Vector
PWM Controller for Voltage-Fed
Inverter Induction Motor Drive
Joao O. P. Pinto, Student Member, IEEE, Bimal K. Bose, Life Fellow, IEEE,
Luiz Eduardo Borges da Silva, Member, IEEE, and Marian P. Kazmierkowski, Fellow, IEEE
Abstract—A neural-network-based implementation of
space-vector modulation (SVM) of a voltage-fed inverter has been
proposed in this paper that fully covers the undermodulation and
overmodulation regions linearly extending operation smoothly
up to square wave. A neural network has the advantage of very
fast implementation of an SVM algorithm that can increase the
converter switching frequency, particularly when a dedicated application-specific integrated circuit chip is used in the modulator.
The scheme has been fully implemented and extensively evaluated
in a V/Hz-controlled 5-hp 60-Hz 230-V induction motor drive. The
performances of the drive with artificial-neural-network-based
SVM are excellent. The scheme can be easily extended to a
vector-controlled drive.
Index Terms—Drive, induction motor, neural network,
pulsewidth modulation.
I. INTRODUCTION
S
PACE-VECTOR modulation (SVM) has recently grown as
a very popular pulsewidth modulation (PWM) method for
voltage-fed converter ac drives because of its superior harmonic
quality and extended linear range of operation [1]. However,
a difficulty of SVM is that it requires complex online computation that usually limits its operation up to several kilohertz
of switching frequency. Of course, switching frequency can be
extended by using a high-speed digital signal processor (DSP)
and simplifying computations with the help of lookup tables.
Lookup tables, unless very large, tend to reduce the pulsewidth
resolution. Very recently, a novel decision-tree-based SVM
technique has been proposed [2] which extends the switching
frequency up to 16 kHz in a low-inductance servo motor
Paper IPCSD 00–036, presented at the 1999 Industry Applications Society
Annual Meeting, Phoenix, AZ, October 3–7, and approved for publication in
the IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS by the Industrial Drives
Committee of the IEEE Industry Applications Society. Manuscript submitted
for review October 15, 1999 and released for publication July 17, 2000. This
work was supported in part by the U.S.–Poland MSC Joint Fund II of Poland
and CAPES of Brazil.
J. O. P. Pinto and B. K. Bose are with the Department of Electrical Engineering, University of Tennessee, Knoxville, TN 37996-2100 USA (e-mail:
jpinto@utk.edu; bbose@utk.edu).
L. E. Borges is with the Department of Electrical Engineering, University of Tennessee, Knoxville, TN 37996-2100 USA, on leave from the
Escola Federal de Engenharia de Itajubá, Itajubá 37500-000, Brazil (e-mail:
leborges@iee.efei.rmg.br).
M. P. Kazmierkowski is with the Institute of Control and Industrial Electronics, Warsaw University of Technology, Warsaw 00-662, Poland (e-mail:
kazmierkowski@isep.pw.edu.pl).
Publisher Item Identifier S 0093-9994(00)10421-9.
drive. Power semiconductor switching speed has improved
dramatically in recent years. Modern ultrafast insulated gate
bipolar transistors (IGBTs) demand switching frequency as
high as 50 kHz. The DSP-based SVM practically fails in this
region where artificial-neural-network (ANN)-based SVM can
possibly take over.
The application of ANNs is recently growing in power
electronic systems. A feedforward ANN implements nonlinear
input–output mapping. The computational delay of this mapping becomes negligible if parallel architecture of the network
is implemented by an application-specific integrated circuit
(ASIC) chip. A feedforward carrier-based PWM technique,
such as SVM, can also be looked upon as a nonlinear mapping
phenomenon where the command phase voltages are sampled
at the input and the corresponding pulsewidth patterns are
established at the output. Therefore, it appears logical that a
backpropagation-type ANN which has high computational
capability can implement an SVM algorithm. The ANN can
be conveniently trained offline with the data generated by
calculation of the SVM algorithm. Note that the ANN has
inherent learning capability that can give improved precision
by interpolation unlike the standard lookup table method.
In the past, a neural network was generally considered for
implementation of instantaneous current control PWM [4]–[7]
where the sinusoidal phase current commands were compared
with the respective feedback currents, and the resulting loop
errors generated the pulsewidth signals through a feedforward
ANN. Current control PWM does not give optimum performance. Besides, it cannot be used in voltage control PWM in
drives which are so widely used in industry. The ANN-based
SVM of a voltage-fed inverter was first proposed in [8], where
the authors considered the undermodulation region only. They
suggested the use of competitive neural network architecture
to identify the inverter switching states and the corresponding
voltage magnitudes for the impressed command voltage.
This paper describes feedforward ANN-based SVM that
fully covers the undermodulation and overmodulation regions
extending operation up to square wave. In the beginning, the
SVM theory has been briefly reviewed with mathematical
analysis for the overmodulation range. Then, equations for
algorithms of undermodulation and overmodulation (Mode-1
and Mode-2) have been developed in detail and simplified so
that the undermodulation algorithm can be extended in the
overmodulation range by modification of a voltage scale factor.
A backpropagation-type feedforward ANN is trained offline
0093–9994/00$10.00 © 2000 IEEE
PINTO et al.: NEURAL-NETWORK-BASED SPACE-VECTOR PWM CONTROLLER
1629
Fig. 1. V/Hz control of induction motor showing neural-network-based implementation.
(a)
(b)
(c)
Fig. 2. Voltage trajectories in undermodulation and overmodulation regions and the corresponding phase voltages. (a) Undermodulation. (b) Overmodulation
Mode-1. (c) Overmodulation Mode-2.
with the data generated by this simple algorithm, and then
extensively evaluated on a V/Hz-controlled induction motor
drive.
II. SPACE-VECTOR PWM IN UNDERMODULATION
OVERMODULATION REGIONS
AND
The SVM theory, in general, has been well discussed in the
literature, and a number of authors [1], [10]–[13] have described
its operation in the overmodulation region. The principles are
briefly reviewed here, adding analysis in the overmodulation
region with the goal of its ANN implementation. The principles described in [10] will be generally accepted. Fig. 1 shows
the block diagram of an open-loop V/Hz-controlled induction
motor drive incorporating the proposed ANN-based SVM conis generated from the
troller. The command voltage
frequency or speed command, and the angle command is ob-
tained by integrating the frequency, as shown. The output of the
modulator generates the PWM patterns for the inverter switches.
For a vector-controlled drive with synchronous current control,
both
and
signals are present and, therefore, the SVM
controller input signals can be modified as follows:
(1)
(2)
Fig. 2 explains SVM operation in the undermodulation and
overmodulation regions, and the corresponding line-to-neutral
average phase voltage ( ) waves are shown in the lower part.
It is well known that the operation of a three-phase two-level
voltage-fed inverter is characterized by eight switching states
] are active to form a hexagon,
of which six [
and
] are zero states that lie at the
and two [
1630
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 6, NOVEMBER/DECEMBER 2000
origin. The operation in undermodulation and overmodulation
is determined by the modulation factor which is defined as
(3)
equations for the four voltage segments in the first quarter-cycle
[12] can be given as
segment 1:
(9)
is the magnitude of command or reference voltage
where
is the peak value (
) of square (or six step)
vector and
voltage wave. The modulation factor varies between 0–1.
A. Undermodulation (
segment 2:
(10)
segment 3:
(11)
)
In the undermodulation or linear region, shown in Fig. 2(a),
remains within the hexagon.
the rotating reference voltage
describes the inThe mode ends at the upper limit when
. The SVM strategy in this region is
scribed circle at
based on generating three consecutive switching voltage vectors
in a sampling period ( ) such that the average output voltage
matches with the reference voltage. The equations for effective
time of the inverter switching states can be given as
(4)
(5)
(6)
segment 4:
(12)
,
is the slope of linear segment
where
, and
is the modified reference voltage. The
1,
can be defined as a function of crossover angle by
voltage
) as
equating (9) and (10) at angle (
(13)
Because of quarter-cycle symmetry, the fundamental output
voltage ( ) expression can be written from (9) to (12) as
where
time of switching vector that lags
;
time of switching vector that leads
;
time of zero-switching vector;
, sampling time (
switching frequency);
angle of
in a 60 sector;
.
The times are distributed to generate symmetrical PWM pulses
(see Fig. 6).
B. Overmodulation Mode-1 (
)
The overmodulation or nonlinear operation starts when the
exceeds the hexagon boundary. In overreference voltage
crosses the hexagon
modulation Mode-1, shown in Fig. 2(b),
at two points in each sector. To compensate loss of fundamental
voltage, i.e., to track with the reference voltage, a modified reference voltage trajectory is selected that remains partly on the
hexagon and partly on a circle. The circular part of the trajectory
and crosses the hexagon at
has extended radius
angle , as shown in the figure. Note that (4)–(6) remain valid for
is replaced by
).
the circular part of the trajectory (except
However, on the hexagon trajectory, time vanishes, giving
and expressions as
(7)
(14)
A computer program is written to solve crossover angle as
from (3) and (9) to (14).
a function of modulation factor
Fig. 3(a) shows the plot of this relation. Mode-1 ends when angle
at
, i.e., the trajectory is on the hexagon giving
voltage wave.
only linear segments of the
C. Overmodulation Mode-2 (
)
increases
In overmodulation Mode-2, the reference vector
further. Therefore, the actual trajectory is modified so that
the output fundamental voltage matches that of the reference
voltage, as discussed previously. The operation in this region,
as explained in Fig. 2(c), is characterized by partly holding the
,
modified vector at the hexagon corner for holding angle
and partly tracking the hexagon sides in every sector. During
remains constant, whereas
holding angle, the magnitude of
during hexagon tracking, the voltage changes almost linearly,
as shown in the lower part of Fig. 2(c). At the end of Mode-2,
the linear segments vanish, giving six-step or square-wave
operation when the modified vector is held at hexagon corners
. An expression for modified
angle
for 60 , i.e.,
( ) in Mode-2 can be given as [10]
(8)
voltage wave is given by approximate linear segments
The
for the hexagon trajectory and sinusoidal segments for the circular trajectory, as shown in the lower part of Fig. 2(b). The
(15)
PINTO et al.: NEURAL-NETWORK-BASED SPACE-VECTOR PWM CONTROLLER
1631
Fig. 4. Turn-on time (T
sectors.
) of phase A as function of angle in different
Fig. 3. (a) Modulation factor (m) relation with crossover phase angle ( ) in
overmodulation Mode-1. (b) Modulation factor (m) relation with holding angle
( ) in overmodulation Mode-2.
For the phase voltage ( ) wave, the equations for the four
segments in the first quarter-cycle [12] can be given as
segment 1:
(16)
segment 2:
(17)
segment 3:
(18)
segment 4:
(19)
Fig. 5.
where
f (V
)
0V
relation in undermodulation and overmodulation regions.
For the output fundamental voltage to match with the reference
and modulation
voltage, a relation between holding angle
factor can be derived from (3) and (16) to (20). The relation
is plotted in Fig. 3(b).
III. NEURAL-NETWORK-BASED SPACE-VECTOR PWM
Again, because of quarter-cycle symmetry, the fundamental
expression can be given as
voltage
The SVM algorithm described in the previous section will
be utilized to generate training data for ANN-based SVM. In
the timer-based method, discussed in this section, the algorithm
will be somewhat modified and then simplified.
A. Undermodulation Region
(20)
Consider typical undermodulation PWM waveforms shown
of Fig. 2(a). For any
in Fig. 6 which are drawn for sector
angle, the timing intervals , , and are given by (4)–(6),
respectively. Similar timing intervals can be calculated for all
1632
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 6, NOVEMBER/DECEMBER 2000
Fig. 6. Neural network topology (2-20-4) for PWM wave synthesis.
six sectors and, correspondingly, the phase A turn-on time can
be expressed as
which is defined as turn-on pulsewidth function at unit amplitude (
). In the undermodulation region, the scale
(see Fig. 5). Equation (24) is
factor is linear, i.e.,
plotted for all sectors in Fig. 4. For phases B and C, the
curves are similar but respectively phase shifted by 120 . At
the upper limit of undermodulation, the curve will reach the
and 0, as indicated in the figure. At the
clamping levels
which makes
always equal to
.
lower limit,
B. Overmodulation Mode-1
(21)
Because of symmetry, the corresponding turn-off time is given
as
As discussed before, the modified voltage trajectory in
Mode-1 is characterized by a part on the hexagon and a part
on a circle. The circular part of the trajectory gives
the same as (24) except
which is given by (13).
Note that the scale factor is a nonlinear function of angle, i.e.,
for
modulation factor . On the hexagon trajectory,
the different sectors can be calculated as
(22)
Equation (21) can be written in the general form
(23)
where
is the voltage amplitude scale factor and
(24)
(25)
and expressions are given by (7) and (8), respecwhere
is plotted in Fig. 4 for all the
tively. The typical time
PINTO et al.: NEURAL-NETWORK-BASED SPACE-VECTOR PWM CONTROLLER
1633
TABLE I
DRIVE SYSTEM PARAMETERS
DC-link voltage
(Vd )
Sampling time (Ts )
Induction motor
300 V
50 s (and 100 s)
5 hp 230 V four pole, Reliance Electric
NEMA-Class B
Frequency range: 0–60 Hz
Power factor (full load): 0.85
Efficiency (full load): 86%
Stator resistance (Rs ): 0.5814 Rotor resistance (Rr ): 0.4165 Stator leakage inductance (Lls ): 3.479 mH
Rotor leakage inductance (Llr ): 4.15 mH
Magnetizing inductance (Lm ): 78.25 mH
Rotor inertia (J): 0.1 kg1m2
Fan load (TL = K!r2 ): K = 8.25 2 105
Fig. 8. Overmodulation Mode-1 phase current waves at 56 Hz. (a)
Neural-network-based SVM. (b) Conventional DSP-based SVM.
Fig. 7. Undermodulation phase current waves at 45
Neural-network-based SVM. (b) Conventional DSP-based SVM.
Hz.
(a)
sectors which indicates clamping at 0 and
. Note that, as
the modulation factor increases, the angle decreases and the
greater amount of the trajectory remains on the hexagon. This
1, 3, 4, and 6, and
increases the clamping regions in sectors
2
steepens the slope of the nearly straightline segments in
and 5. At the end of Mode-1 that corresponds to fully hexagon
trajectory, the sectors 1, 3, 4, and 6 remain fully clamped.
C. Overmodulation Mode-2
The time
cept the angle
expression in Mode-2 is given by (25), exis replaced by
which is given by (15).
Fig. 9. Overmodulation Mode-2 phase current waves at 59 Hz. (a)
Neural-network-based SVM. (b) Conventional DSP-based SVM.
The plot of the
curve in Fig. 4 indicates full clamping
in sectors 1, 3, 4, and 6 and partial clamping in sectors 2 and 5.
At the end of Mode-2, clamping occurs in all the sectors and the
curve is given by a square wave clamped at the levels
and 0. This describes the square-wave operation of the inverter.
1634
Fig. 10.
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 6, NOVEMBER/DECEMBER 2000
V/Hz-controlled drive performance covering all the modes. (a) Neural-network-based SVM. (b) Conventional DSP-based SVM.
D. Simplification of Overmodulation Region
curves shown in Fig. 4 for undermoduSolving the
lation and overmodulation Mode-1 and Mode-2 by the respective equation sets for generating the neural network training data
gives a considerable amount of complexity. It can be shown
that the undermodulation curve can be expanded to Mode-1 and
, as shown in
Mode-2 by using a nonlinear scale factor
Fig. 5, in order to get linear transfer characteristics in the whole
range. The idea of this simplification is analogous to the sinusoidal PWM overmodulation principle where the modulating
voltage is magnified. In the undermodulation region,
, as explained before. This means that the
curve can be
to get
. In Mode-1,
plotted in Fig. 4 and multiplied by
is determined by using (13) and Fig. 3(a). In the proposed
simplification, it can be shown that the curves are prone to some
error only in sectors 2 and 3. Rigorous evaluation indicates that
the average error is less than 0.28% by this simplification. For
Mode-2 operation, the scale factor is determined by expanding
the undermodulation curve and matching with the curve established by analysis. The resulting average error is below 0.5%.
increases steeply in Mode-2 until it is
As shown in Fig. 5,
theoretically infinity at square wave.
E. Neural Network—Angle and Amplitude Subnets
and simplification
The derivation of turn-on time
of modes discussed in the previous section permit ANN-based
SVM implementation using two separate subnets: angle subnet
and amplitude subnet. Fig. 6 shows the proposed ANN topology
including the timer section. The subnets use a multilayer perception-type network with sigmoidal-type transfer function.
The bias is not shown in the figure. The composite network
uses two neurons at the input, 20 neurons in the hidden layer,
and four output neurons. The input signal to the angle subnet
is angle which is normalized and then pulsewidth functions
at unit amplitude are solved (or mapped) at the output for the
three phases, as indicated. The amplitude subnet implements
function of Fig. 5. The digital words corresponding
the
to turn-on time are generated by multiplying the angle subnet
output with that of the amplitude subnet and then adding
bias signal, as shown. The PWM signals are then
the
generated using a single timer. The angle subnet is trained
with an angle interval of 2.16 in the range of 0 –360 . The
backpropagation algorithm in the MATLAB Neural Network
Toolbox is used for the training. The angle subnet takes 4800
epochs for training with an error below 0.5%. The amplitude
subnet was trained with data at 1.0 V increments in the range
of 0–191 V. It takes 1750 epochs for the training with an error
below 1.0%. The training data size and the corresponding
training time are, thus, reasonably small. It is important to
note that, due to learning or interpolation capability, both the
subnets will operate with higher signal resolution. Two values
of sample time ( ) are used in the project, i.e., 50 s that
corresponds to 20-kHz switching frequency and 100 s that
corresponds to 10-kHz switching frequency. The network is
solved every sampling time to establish pulsewidth signals
at the output. Note that the network works well in the full
frequency range, including dc (zero frequency) condition.
IV. PERFORMANCE EVALUATION OF INDIRECT METHOD
As mentioned before, the ANN-based SVM was implemented for the PWM controller and the drive system
performance was evaluated. Once the network was trained
and tested in the full range of voltage and frequency using
s (10 kHz) and 50 s (20
sampling time
PINTO et al.: NEURAL-NETWORK-BASED SPACE-VECTOR PWM CONTROLLER
1635
kHz), it was integrated with a V/Hz-controlled induction
motor drive (Fig. 1). Table I shows the parameters of the
drive system. The estimated computation time of the PWM
controller is 40 s with a TMS320C30-type DSP. Unfortunately, no suitable commercial ASIC chip is yet available
[14] to implement the controller economically. The drive
performance was evaluated in detail in undermodulation and
overmodulation regions and compared with the performance
of a conventional SVM-controlled drive. They were found
to correlate very well. Fig. 7(a) shows the undermodulation
phase current waves at 45-Hz frequency, and Fig. 7(b) shows
the corresponding waves with the conventional DSP-based
modulator. The distortion of both cases was analyzed by
fast Fourier transform (FFT) and found to be 0.025% and
0.023%, respectively (note that a six-step wave has 4.5%
distortion). The training algorithm simplification and finite
training error of the ANN did not seem to contribute much
performance deterioration. Fig. 8 shows the corresponding
comparison of current waves in overmodulation Mode-1 at
56 Hz. Although the waveforms look practically identical,
the distortion of the ANN-based SVM was 0.176%, and
for the conventional SVM it was 0.091%. Fig. 9 shows the
current waves in overmodulation Mode-2 at 59 Hz. The corresponding distortions were 0.858% and 0.278%. The drive
was then accelerated in steps with ramp frequency command
to cover operation in all the modes, and performance was
found to be excellent. Fig. 10 shows the comparison of
performance with fan load (see Table I) where speed
is indicated in electrical radians per second. The drive
)
goes into overmodulation region at 344 rad/s (
). The
and field-weakening region at 377 rad/s (
drive is accelerated with ramp speed command ( ) within
the constraint of rated torque (20 N m). The actual speed
follows the command except the difference of positive slip
frequency. Some amount of oscillatory behavior is evident
in open-loop control, particularly at low speed. As the
speed enters into the overmodulation and field-weakening
regions, the ripple in the developed torque due to distortion
in the current wave is evident. The ANN-based SVM with
s was similarly evaluated in the drive and the
performance correlated equally well. Currently, the PWM
controller is being used successfully in a stator-flux-oriented
vector-controlled induction motor drive.
a V/Hz-controlled induction motor drive, and gives excellent
performance. The PWM controller is currently being used
in a stator-flux-oriented vector-controlled induction motor
drive. The ANN-based SVM can give higher switching
frequency, which is not possible by conventional DSP-based
SVM. The switching frequency can be easily extended up to
50 kHz if the ANN is implemented by a dedicated hardware
ASIC chip.
REFERENCES
[1] J. Holtz, “Pulse width modulation for electric power conversion,” Proc.
IEEE, vol. 82, pp. 1194–1214, Aug. 1994.
[2] J. O. Krah and J. Holtz, “High performance current regulation for low
inductance servo motors,” in Conf. Rec. IEEE-IAS Annu. Meeting, 1998,
pp. 490–499.
[3] B. K. Bose, Power Electronics and AC Drives. Englewood Cliffs, NJ:
Prentice-Hall, 1986.
[4] F. Harashima et al., “Applications of neural networks to power converter
control,” in Conf. Rec. IEEE-IAS Annu. Meeting, 1989, pp. 1086–1091.
[5] M. R. Buhl and R. D. Lorenz, “Design and implementation of neural
networks for digital current regulation of inverter drives,” in Conf. Rec.
IEEE-IAS Annu. Meeting, 1991, pp. 415–423.
[6] J. W. Song, K. C. Lee, K. B. Cho, and J. S. Won, “An adaptive learning
current controller for field-oriented controlled induction motor by neural
network,” in Proc. IEEE IECON’91, 1991, pp. 469–474.
[7] M. P. Kazmierkowski, D. L. Sobczuk, and M. A. Dzieniakowski,
“Neural network current control of VS-PWM inverters,” in Proc.
EPE’95, 1995, pp. 1415–1420.
[8] A. Bakhshai, J. Espinoza, G. Joos, and H. Jin, “A combined artificial
neural network and DSP approach to the implementation of space vector
modulation techniques,” in Conf. Rec. IEEE-IAS Annu. Meeting, 1996,
pp. 934–940.
[9] H. W. Van Der Broeck, H. C. Skudelny, and G. Stanke, “Analysis and
realization of a pulse width modulator based on voltage space vectors,”
IEEE Trans. Ind. Applicat., vol. 24, pp. 142–150, Jan./Feb. 1988.
[10] J. Holtz, W. Lotzkat, and M. Khambadkone, “On continuous control
of PWM inverters in the overmodulation range including the six-step
mode,” IEEE Trans. Power Electron., vol. 8, pp. 546–553, Oct. 1993.
[11] S. Bolognani and M. Ziglitti, “Novel digital continuous control of SVM
inverters in the overmodulation range,” IEEE Trans. Ind. Applicat., vol.
33, pp. 525–530, Mar./Apr. 1997.
[12] J. Kim and S. Sul, “A novel voltage modulation technique of the space
vector PWM,” in Proc. Int. Power Electronics, Yokohama, Japan, Apr.
1995, pp. 742–747.
[13] D. C. Lee and G. M. Lee, “A novel overmodulation technique for space
vector PWM inverters,” IEEE Trans. Power Electron., vol. 13, pp.
1144–1151, Nov. 1998.
[14] L. M. Reynery, “Neuro-fuzzy hardware: design, development and performance,” in Proc. IEEE-FEPPCON III, Kruger National Ppark, South
Africa, July 1998, pp. 233–241.
[15] B. K. Bose, Ed., Power Electronics and Variable Frequency
Drives. Piscataway, NJ: IEEE Press, 1997.
V. CONCLUSION
A neural-network-based space-vector modulator has been
described that operates very well in undermodulation as well
as in overmodulation (Mode-1 and Mode-2) regions. The
digital words corresponding to turn-on time are generated by
the ANN and then converted to pulsewidths through a single
timer. The scheme uses a backpropagation-type feedforward
network. The training data and training time are reasonably
small. The method can operate from dc (zero frequency).
The scheme has been fully implemented and evaluated with
Joao O. P. Pinto (S’97) received the B.S. degree
from the Universidade Estadual Paulista, Ilha
Solteira, Brazil, and the M.S. degree from the Universidade Federal de Uberlândia, Uberlândia, Brazil,
in 1990 and 1993, respectively. He is currently
working toward the Ph.D. degree at the University
of Tennessee, Knoxville.
He has been an Assistant Professor at the Universidade Federal do Mato Grosso do Sul, Campo
Grande, Brazil, since 1994. His research interests
include neural networks, fuzzy logic, genetic
algorithms, and wavelet applications to power electronics, PWM techniques,
drives, and machine control.
1636
IEEE TRANSACTIONS ON INDUSTRY APPLICATIONS, VOL. 36, NO. 6, NOVEMBER/DECEMBER 2000
Bimal K. Bose (S’59–M’60–SM’78–F’89–LF’96)
received the B.E. degree from Calcutta University
(Bengal Engineering College), Calcutta, India, the
M.S. degree from the University of Wisconsin,
Madison, and the Ph.D. degree from Calcutta
University in 1956, 1960, and 1966, respectively.
He currently holds the Condra Chair of Excellence
in Power Electronics at the University of Tennessee,
Knoxville, where he has been responsible for
organizing the power electronics program for the
last 13 years. He is also the Distinguished Scientist
(formerly Chief Scientist) of the EPRI-Power Electronics Applications Center,
Knoxville, TN, Honorary Professor of Shanghai University, China University
of Mining and Technology, and Xian Mining Institute (also Honorary Director
of the Electrical Engineering Institute), China, and Senior Adviser of the
Beijing Power Electronics Research and Development Center, Beijing, China.
Early in his career, he served as a Faculty Member at Calcutta University
for 11 years. In 1971, he joined Rensselaer Polytechnic Institute, Troy, NY,
as an Associate Professor of Electrical Engineering. In 1976, he joined
General Electric Corporate Research and Development, Schenectady, NY, as
Research Engineer and served there for 11 years. He has been a Consultant
to more than ten industries. His research interests spread over the whole
spectrum of power electronics, and specifically include power converters, ac
drives, microcomputer control, EV drives, and expert system, fuzzy logic,
and neural-network-based control of power electronics and drive systems.
He has authored more than 150 published papers and is the holder of 21
U.S. patents. He is the author/editor of five books: Power Electronics and
AC Drives (Englewood Cliffs, NJ: Prentice–Hall, 1986), Adjustable Speed
AC Drive Systems (New York: IEEE Press, 1981), Microcomputer Control
of Power Electronics and Drives (New York: IEEE Press, 1987), Modern
Power Electronics (New York: IEEE Press, 1992), and Power Electronics and
Variable Frequency Drives (Piscataway, NJ: IEEE Press, 1997). Another book,
Modern Power Electronics and AC Drives, is to be published by Prentice-Hall.
The book Power Electronics and AC Drives has been translated into Japanese,
Chinese, and Korean.
Dr. Bose was the Guest Editor of the PROCEEDINGS OF THE IEEE Special
Issue on Power Electronics and Motion Control (August 1994). He has received
the GE Publication Award, Silver Patent Medal, and a number of IEEE Prize
Paper Awards. He is listed in Marquis’ Who’s Who in America. He has served
the IEEE in various capacities, including Chairman of the IEEE Industrial
Electronics Society (IES) Power Electronics Council, Associate Editor of
IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, IEEE-IECON Power
Electronics Chairman, Chairman of the IEEE Industry Applications Society
(IAS) Industrial Power Converter Committee, IAS member in the Neural
Networks Council, and various other national and international professional
committees. He has been a member of the Editorial Board of the PROCEEDINGS
OF THE IEEE since 1995 and has served a large number of national and
international conferences. He has served as a Distinguished Lecturer in the
IAS and IES. For his research contributions, he was awarded the Premchand
Roychand Scholarship and the Mouat Gold Medal by Calcutta University in
1968 and 1970, respectively. He received the IAS Outstanding Achievement
Award in 1993, the IES Eugene Mittelmann Award in 1994, IEEE Region 3
Outstanding Engineer Award in 1994, IEEE Lamme Gold Medal in 1996, IEEE
Continuing Education Award in 1997, and IEEE Millenium Medal in 2000.
Luiz Eduardo Borges da Silva (S’84–M’89) was
born in Passa Quatro, Brazil. He received the B.S.
degree in electrical engineering and the M.S. degree
from the Escola Federal de Engenharia de Itajubá,
Itajubá, Brazil, and the Ph.D. degree in power electronics from the Ecole Polytechnique de Montreal,
Montreal, PQ, Canada, in 1977, 1982, and 1988, respectively.
From 1988 to 1992, he was the Head of the Electronics Department, Escola Federal de Engenharia
de Itajubá. During 1998, he held a post-doctoral
position in the Department of Electrical Engineering, University of Tennessee,
Knoxville. He teaches power electronics and digital control systems at the
Escola Federal de Engenharia de Itajubá, where he also works with the Power
Electronics group. He is currently on leave at the University of Tennessee,
Knoxville. His main research interests are applications of artificial intelligence
in power electronics and signal processing.
Marian P. Kazmierkowski (M’89–SM’91–F’98)
received the M.Sc., Ph.D., and Dr. Sc. degrees in
electrical engineering from the Institute of Control
and Industrial Electronics, Warsaw University of
Technology, Warsaw, Poland, in 1968, 1972 and
1981, respectively.
From 1967 to 1969, he was with the Industrial
Research Institute of Electrotechnics, Warsaw,
Poland, and from 1969 to 1980, he was with the
Institute of Control and Industrial Electronics,
Warsaw University of Technology, as an Assistant
Professor. From 1980 to 1983, he was with RWTH Aachen, Aachen, West
Germany, as an Alexander von Humboldt Fellow. During 1986–1987, he was
a Visiting Professor at NTH Trondheim, Trondheim, Norway. Since 1987, he
has been a Professor and Director of the Institute of Control and Industrial
Electronics, Warsaw University of Technology. He was a Visiting Professor
at the University of Minnesota, Minneapolis, in 1990, at Aalborg University,
Aalborg East, Denmark, in 1990 and 1995, and at the University of Padova,
Padova, Italy, in 1993. He is also currently a Coordinating Professor in the
International Danfoss Professor Program 1997–2000, Aalborg University.
Since 1996, he has served as an elected member of the State Committee
for Scientific Research in Poland. He is engaged in research and theoretical
work on electrical drive control and industrial electronics. He is the author or
coauthor of more than 140 technical papers and reports, as well as 11 books
and textbooks. His latest book, with Dr. H. Tunia, is Automatic Control of
Converter-Fed Drives (Amsterdam, The Nederlands: Elsevier, 1994).
Dr. Kazmierkowski was Chairman of the 1996 IEEE International Symposium on Industrial Electronics held in Warsaw, Poland. He is an Associate Editor of the IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, a Vice President
of the IEEE Industrial Electronics Society, and the IEEE Industrial Electronics
and IEEE Power Electronics Societies Joint Chapter Chairman of the Poland
Section of the IEEE.